Where will artificial intelligence take us? Several recent investigations review the latest advances in artificial intelligence and the prospects for future progress of this fascinating but also disturbing unexplored territory into which human civilization has begun to enter.
These studies have been jointly published in a special issue of the academic journal Science.
Racial variables predicted by artificial intelligence from medical imaging pose risks and opportunities for studying health disparities, James Zou and colleagues say in an analysis.
Hundreds of AI-assisted medical devices are currently used in various medical tasks, such as health risk assessment and diagnosis of diseases such as cancer. Some studies have shown that artificial intelligence models can infer racial variables, albeit in rudimentary and simplistic categories, directly from medical images such as chest X-rays and cardiac ultrasounds, despite the fact that there are no known human-readable racial correlations in these images.
“Although racial variables are not a generally significant category in medicine, the ability of artificial intelligence to predict racial variables from medical images could be useful in monitoring disparity in care and ensuring algorithms work well across diverse populations,” Zou and colleagues write.
In another analysis, Matthew DeCamp (University of Colorado, USA) and Charlotta Lindvall highlight how examining bias in AI and healthcare has tended to remove bias from AI data sets, analyses, or development teams. However, DeCamp and Lindvall argue that biases in the way that doctors and patients use AI-based algorithms will also need to be reduced, which could be more complex than reducing biases in the algorithms themselves.
Artificial intelligence technologies also offer great promise for expanding our understanding of animal behaviors. In a study, Christian Rutz (University of St Andrews in the UK) and his colleagues review how machine learning (a form of artificial intelligence) methods are being used to decode animal communication systems. Understanding how animals communicate presents a number of challenges: Animals use a wide range of communicative adaptations that encompass visual, acoustic, tactile, chemical, and electrical signals, often in ways that are beyond the perceptual abilities of humans. Here, Rutz and colleagues review the ways in which increasingly powerful machine learning tools are being used to reveal previously hidden complexity in animal communication behavior and provide insights that could lead to potential benefits for animal welfare and conservation. “It is essential that future advances be used for the benefit of the animals being studied,” Rutz and his colleagues write.
Peter Wurman’s team (Sony Research) highlights how games provide controlled opportunities to isolate and practice multiple problem-solving skills that are more broadly transferable to real-world applications, making them valuable training grounds for intelligent machines. While AI’s recent dominance in classic strategy games is largely complete, Wurman and his colleagues argue that video games pose new kinds of challenges for AI to conquer. Progress in these areas will represent a substantial step towards much more capable and flexible artificial intelligence systems that operate in the physical world.
The general public, scientists, and technologists have widely and rapidly adopted generative artificial intelligence – a type of artificial intelligence technology capable of producing a wide variety of content, including images, video, audio, and text. However, a growing number of professional artists, including writers and musicians, have raised objections to the use of their creations as training data for these systems. Pamela Samuelson (University of California at Berkeley, USA) highlights this emerging conflict and discusses how several copyright lawsuits currently underway in the United States could have substantial implications for the future of generative AI systems. If the plaintiffs’ view prevails in these cases, the only materials with which generative artificial intelligence systems could legally be trained would be public domain or licensed works, which would affect all users of this technology, including scientific research.
For their part, Ajay Agrawal and colleagues discuss how automating tasks through innovations in artificial intelligence could reverse current trends of rising income inequality. Given the rapid development of artificial intelligence technologies that enable the automation of cognitive and creative endeavors once reserved for humans with specialized training and experience, some economists have raised concerns that artificial intelligence has the potential to substantially disrupt the labor market and further increase inequality, albeit with little benefit to productivity and living standards. Here, Agrawal and his colleagues argue that by considering how tasks can be automated, AI developers could create tools that improve the overall productivity of workers. Furthermore, AI automation could also reduce income inequality by delivering innovations that enable lower-paid and less-skilled workers to perform at levels that would previously require specialized training.
Will artificial intelligence change the world forever? (Illustration: Jorge Munnshe for NCYT from Amazings)
Felix Wong and his colleagues discuss how advances in artificial intelligence are empowering medical and biotech research in the fight against infectious diseases. According to Wong’s team, artificial intelligence technologies such as machine learning have led to rapid advances in anti-infective drug discovery, as well as our understanding of the biology of infection and the development of new diagnostics. Other apps could also improve our ability to forecast and control infectious disease outbreaks and pandemics.
Bing Huang (University of Vienna in Austria) and colleagues discuss the crucial role that “Functional Theory of Fate” – fundamental in chemical and materials science due to its relatively high predictive power – has played in the development of machine learning-based models used to navigate the space of chemical compounds. Huang and his colleagues argue that continued advances in this space pave the way for software control solutions capable of routinely handling unusual chemistries and formulations within self-managed laboratories.
Finally, a series of small reviews by various authors highlight the applications of artificial intelligence in advanced medical robots. Artificial intelligence technologies used in these devices, including computer vision, medical image analysis, very high-precision manipulation (beyond human capabilities), and machine learning, could enable autonomous robots to take diagnostic images as well as assist in complex surgical procedures. Additionally, AI in wearable rehabilitation devices and advanced prosthetics could enable more personalized patient care and even AI-powered prosthetics that operate seamlessly with the human user. (Source: AAAS)